Kelvin Wong

Alexander Cui, Sergio Casas, Kelvin Wong et al. · utoronto

The task of motion forecasting is critical for self-driving vehicles (SDVs) to be able to plan a safe maneuver. Towards this goal, modern approaches reason about the map, the agents' past trajectories and their interactions in order to produce accurate forecasts. The predominant approach has been to encode the map and other agents in the reference frame of each target agent. However, this approach is computationally expensive for multi-agent prediction as inference needs to be run for each agent. To tackle the scaling challenge, the solution thus far has been to encode all agents and the map in a shared coordinate frame (e.g., the SDV frame). However, this is sample inefficient and vulnerable to domain shift (e.g., when the SDV visits uncommon states). In contrast, in this paper, we propose an efficient shared encoding for all agents and the map without sacrificing accuracy or generalization. Towards this goal, we leverage pair-wise relative positional encodings to represent geometric relationships between the agents and the map elements in a heterogeneous spatial graph. This parameterization allows us to be invariant to scene viewpoint, and save online computation by re-using map embeddings computed offline. Our decoder is also viewpoint agnostic, predicting agent goals on the lane graph to enable diverse and context-aware multimodal prediction. We demonstrate the effectiveness of our approach on the urban Argoverse 2 benchmark as well as a novel highway dataset.

Haomiao Ni, Yuan Xue, Kelvin Wong et al.

Accurate infarct segmentation in non-contrast CT (NCCT) images is a crucial step toward computer-aided acute ischemic stroke (AIS) assessment. In clinical practice, bilateral symmetric comparison of brain hemispheres is usually used to locate pathological abnormalities. Recent research has explored asymmetries to assist with AIS segmentation. However, most previous symmetry-based work mixed different types of asymmetries when evaluating their contribution to AIS. In this paper, we propose a novel Asymmetry Disentanglement Network (ADN) to automatically separate pathological asymmetries and intrinsic anatomical asymmetries in NCCTs for more effective and interpretable AIS segmentation. ADN first performs asymmetry disentanglement based on input NCCTs, which produces different types of 3D asymmetry maps. Then a synthetic, intrinsic-asymmetry-compensated and pathology-asymmetry-salient NCCT volume is generated and later used as input to a segmentation network. The training of ADN incorporates domain knowledge and adopts a tissue-type aware regularization loss function to encourage clinically-meaningful pathological asymmetry extraction. Coupled with an unsupervised 3D transformation network, ADN achieves state-of-the-art AIS segmentation performance on a public NCCT dataset. In addition to the superior performance, we believe the learned clinically-interpretable asymmetry maps can also provide insights towards a better understanding of AIS assessment. Our code is available at https://github.com/nihaomiao/MICCAI22_ADN.

Yanglan Ou, Ye Yuan, Xiaolei Huang et al. · cmu

We present a new encoder-decoder Vision Transformer architecture, Patcher, for medical image segmentation. Unlike standard Vision Transformers, it employs Patcher blocks that segment an image into large patches, each of which is further divided into small patches. Transformers are applied to the small patches within a large patch, which constrains the receptive field of each pixel. We intentionally make the large patches overlap to enhance intra-patch communication. The encoder employs a cascade of Patcher blocks with increasing receptive fields to extract features from local to global levels. This design allows Patcher to benefit from both the coarse-to-fine feature extraction common in CNNs and the superior spatial relationship modeling of Transformers. We also propose a new mixture-of-experts (MoE) based decoder, which treats the feature maps from the encoder as experts and selects a suitable set of expert features to predict the label for each pixel. The use of MoE enables better specializations of the expert features and reduces interference between them during inference. Extensive experiments demonstrate that Patcher outperforms state-of-the-art Transformer- and CNN-based approaches significantly on stroke lesion segmentation and polyp segmentation. Code for Patcher is released with publication to facilitate future research.

Zimu Gong, Brian Zhaoning Zhang, Chris Zhang et al.

Safety-critical scenarios are essential for the development of autonomous vehicles (AVs) but are rare in real-world driving data. While simulation offers a way to generate such scenarios, manually designed test cases lack scalability, and adversarial optimization often produces unrealistic behaviors. In this work, we introduce a conditional latent flow matching approach for scalable and realistic safety-critical scenario generation. Our method uses distribution matching to transform nominal scenes into safety-critical rollouts. Furthermore, we demonstrate that incorporating both simulation and real-world data enables our framework to efficiently generate diverse, data-driven scenarios. Experimental results highlight that our approach is able to more consistently and realistically generate novel safety-critical scenarios, making it a valuable tool for training and benchmarking AV systems.

Chris Zhang, James Tu, Lunjun Zhang et al.

Realistic traffic simulation is crucial for developing self-driving software in a safe and scalable manner prior to real-world deployment. Typically, imitation learning (IL) is used to learn human-like traffic agents directly from real-world observations collected offline, but without explicit specification of traffic rules, agents trained from IL alone frequently display unrealistic infractions like collisions and driving off the road. This problem is exacerbated in out-of-distribution and long-tail scenarios. On the other hand, reinforcement learning (RL) can train traffic agents to avoid infractions, but using RL alone results in unhuman-like driving behaviors. We propose Reinforcing Traffic Rules (RTR), a holistic closed-loop learning objective to match expert demonstrations under a traffic compliance constraint, which naturally gives rise to a joint IL + RL approach, obtaining the best of both worlds. Our method learns in closed-loop simulations of both nominal scenarios from real-world datasets as well as procedurally generated long-tail scenarios. Our experiments show that RTR learns more realistic and generalizable traffic simulation policies, achieving significantly better tradeoffs between human-like driving and traffic compliance in both nominal and long-tail scenarios. Moreover, when used as a data generation tool for training prediction models, our learned traffic policy leads to considerably improved downstream prediction metrics compared to baseline traffic agents. For more information, visit the project website: https://waabi.ai/rtr

Chris Zhang, Sourav Biswas, Kelvin Wong et al.

Large-scale data is crucial for learning realistic and capable driving policies. However, it can be impractical to rely on scaling datasets with real data alone. The majority of driving data is uninteresting, and deliberately collecting new long-tail scenarios is expensive and unsafe. We propose asymmetric self-play to scale beyond real data with additional challenging, solvable, and realistic synthetic scenarios. Our approach pairs a teacher that learns to generate scenarios it can solve but the student cannot, with a student that learns to solve them. When applied to traffic simulation, we learn realistic policies with significantly fewer collisions in both nominal and long-tail scenarios. Our policies further zero-shot transfer to generate training data for end-to-end autonomy, significantly outperforming state-of-the-art adversarial approaches, or using real data alone. For more information, visit https://waabi.ai/selfplay .

Scott Xu, Dian Chen, Kelvin Wong et al.

Accurately modeling agent behaviors is an important task in self-driving. It is also a task with many symmetries, such as equivariance to the order of agents and objects in the scene or equivariance to arbitrary roto-translations of the entire scene as a whole; i.e., SE(2)-equivariance. The transformer architecture is a ubiquitous tool for modeling these symmetries. While standard self-attention is inherently permutation equivariant, explicit pairwise relative positional encodings have been the standard for introducing SE(2)-equivariance. However, this approach introduces an additional cost that is quadratic in the number of agents, limiting its scalability to larger scenes and batch sizes. In this work, we propose DriveGATr, a novel transformer-based architecture for agent modeling that achieves SE(2)-equivariance without the computational cost of existing methods. Inspired by recent advances in geometric deep learning, DriveGATr encodes scene elements as multivectors in the 2D projective geometric algebra $\mathbb{R}^*_{2,0,1}$ and processes them with a stack of equivariant transformer blocks. Crucially, DriveGATr models geometric relationships using standard attention between multivectors, eliminating the need for costly explicit pairwise relative positional encodings. Experiments on the Waymo Open Motion Dataset demonstrate that DriveGATr is comparable to the state-of-the-art in traffic simulation and establishes a superior Pareto front for performance vs computational cost.

Frieda Rong, Chris Zhang, Kelvin Wong et al.

Autonomous vehicles (AVs) require extensive testing in simulation, but test case generation for driving scenarios is laborious. The desired scenarios are often out-of-distribution and have precise requirements on interactions with the AV policy under test. Manually programming scenarios allows for precise controllability but is difficult to scale. On the other hand, statistical models can leverage compute and data, but struggle with precise controllability when out-of-distribution. We cast scenario orchestration as a constraint-solving problem and present a language-in, simulation-out scenario orchestrator for closed-loop testing AVs. Our approach leverages foundation model reasoning to translate general, natural language descriptions into a set of constraints as a scenario representation. This then allows us to leverage off the shelf solvers to solve for actor behaviors which meet precise testing intentions in closed-loop. Under a benchmark of carefully crafted and diverse scenario descriptions, our approach greatly outperforms our baselines in orchestration success rate. We further show that our closed-loop approach is especially important for scenarios which require ego-reactive specifications.

Chao Liu, Boxi Chen, Wei Shao et al.

The advent of the Internet of Things (IoT) has brought forth an era of unprecedented connectivity, with an estimated 80 billion smart devices expected to be in operation by the end of 2025. These devices facilitate a multitude of smart applications, enhancing the quality of life and efficiency across various domains. Machine Learning (ML) serves as a crucial technology, not only for analyzing IoT-generated data but also for diverse applications within the IoT ecosystem. For instance, ML finds utility in IoT device recognition, anomaly detection, and even in uncovering malicious activities. This paper embarks on a comprehensive exploration of the security threats arising from ML's integration into various facets of IoT, spanning various attack types including membership inference, adversarial evasion, reconstruction, property inference, model extraction, and poisoning attacks. Unlike previous studies, our work offers a holistic perspective, categorizing threats based on criteria such as adversary models, attack targets, and key security attributes (confidentiality, availability, and integrity). We delve into the underlying techniques of ML attacks in IoT environment, providing a critical evaluation of their mechanisms and impacts. Furthermore, our research thoroughly assesses 65 libraries, both author-contributed and third-party, evaluating their role in safeguarding model and data privacy. We emphasize the availability and usability of these libraries, aiming to arm the community with the necessary tools to bolster their defenses against the evolving threat landscape. Through our comprehensive review and analysis, this paper seeks to contribute to the ongoing discourse on ML-based IoT security, offering valuable insights and practical solutions to secure ML models and data in the rapidly expanding field of artificial intelligence in IoT.

Tongan Cai, Haomiao Ni, Wenchao Ma et al.

Effective stroke triage in emergency settings often relies on clinicians' ability to identify subtle abnormalities in facial muscle coordination. While recent AI models have shown promise in detecting such patterns from patient facial videos, their reliance on real patient data raises significant ethical and privacy challenges -- especially when training robust and generalizable models across institutions. To address these concerns, we propose SafeTriage, a novel method designed to de-identify patient facial videos while preserving essential motion cues crucial for stroke diagnosis. SafeTriage leverages a pretrained video motion transfer (VMT) model to map the motion characteristics of real patient faces onto synthetic identities. This approach retains diagnostically relevant facial dynamics without revealing the patients' identities. To mitigate the distribution shift between normal population pre-training videos and patient population test videos, we introduce a conditional generative model for visual prompt tuning, which adapts the input space of the VMT model to ensure accurate motion transfer without needing to fine-tune the VMT model backbone. Comprehensive evaluation, including quantitative metrics and clinical expert assessments, demonstrates that SafeTriage-produced synthetic videos effectively preserve stroke-relevant facial patterns, enabling reliable AI-based triage. Our evaluations also show that SafeTriage provides robust privacy protection while maintaining diagnostic accuracy, offering a secure and ethically sound foundation for data sharing and AI-driven clinical analysis in neurological disorders.

Tongan Cai, Haomiao Ni, Mingli Yu et al.

In an emergency room (ER) setting, stroke triage or screening is a common challenge. A quick CT is usually done instead of MRI due to MRI's slow throughput and high cost. Clinical tests are commonly referred to during the process, but the misdiagnosis rate remains high. We propose a novel multimodal deep learning framework, DeepStroke, to achieve computer-aided stroke presence assessment by recognizing patterns of minor facial muscles incoordination and speech inability for patients with suspicion of stroke in an acute setting. Our proposed DeepStroke takes one-minute facial video data and audio data readily available during stroke triage for local facial paralysis detection and global speech disorder analysis. Transfer learning was adopted to reduce face-attribute biases and improve generalizability. We leverage a multi-modal lateral fusion to combine the low- and high-level features and provide mutual regularization for joint training. Novel adversarial training is introduced to obtain identity-free and stroke-discriminative features. Experiments on our video-audio dataset with actual ER patients show that DeepStroke outperforms state-of-the-art models and achieves better performance than both a triage team and ER doctors, attaining a 10.94% higher sensitivity and maintaining 7.37% higher accuracy than traditional stroke triage when specificity is aligned. Meanwhile, each assessment can be completed in less than six minutes, demonstrating the framework's great potential for clinical translation.

Yanglan Ou, Ye Yuan, Xiaolei Huang et al.

Diffusion-weighted (DW) magnetic resonance imaging is essential for the diagnosis and treatment of ischemic stroke. DW images (DWIs) are usually acquired in multi-slice settings where lesion areas in two consecutive 2D slices are highly discontinuous due to large slice thickness and sometimes even slice gaps. Therefore, although DWIs contain rich 3D information, they cannot be treated as regular 3D or 2D images. Instead, DWIs are somewhere in-between (or 2.5D) due to the volumetric nature but inter-slice discontinuities. Thus, it is not ideal to apply most existing segmentation methods as they are designed for either 2D or 3D images. To tackle this problem, we propose a new neural network architecture tailored for segmenting highly-discontinuous 2.5D data such as DWIs. Our network, termed LambdaUNet, extends UNet by replacing convolutional layers with our proposed Lambda+ layers. In particular, Lambda+ layers transform both intra-slice and inter-slice context around a pixel into linear functions, called lambdas, which are then applied to the pixel to produce informative 2.5D features. LambdaUNet is simple yet effective in combining sparse inter-slice information from adjacent slices while also capturing dense contextual features within a single slice. Experiments on a unique clinical dataset demonstrate that LambdaUNet outperforms existing 3D/2D image segmentation methods including recent variants of UNet. Code for LambdaUNet is released with the publication to facilitate future research.

Min Bai, Shenlong Wang, Kelvin Wong et al.

In this paper, we introduce a non-parametric memory representation for spatio-temporal segmentation that captures the local space and time around an autonomous vehicle (AV). Our representation has three important properties: (i) it remembers what it has seen in the past, (ii) it reinforces and (iii) forgets its past beliefs based on new evidence. Reinforcing is important as the first time we see an element we might be uncertain, e.g, if the element is heavily occluded or at range. Forgetting is desirable, as otherwise false positives will make the self driving vehicle behave erratically. Our process is informed by 3D reasoning, as occlusion is key to distinguishing between the desire to forget and to remember. We show how our method can be used as an online component to complement static world representations such as HD maps by detecting and remembering changes that should be superimposed on top of this static view due to such events.

Shuhan Tan, Kelvin Wong, Shenlong Wang et al.

We consider the problem of generating realistic traffic scenes automatically. Existing methods typically insert actors into the scene according to a set of hand-crafted heuristics and are limited in their ability to model the true complexity and diversity of real traffic scenes, thus inducing a content gap between synthesized traffic scenes versus real ones. As a result, existing simulators lack the fidelity necessary to train and test self-driving vehicles. To address this limitation, we present SceneGen, a neural autoregressive model of traffic scenes that eschews the need for rules and heuristics. In particular, given the ego-vehicle state and a high definition map of surrounding area, SceneGen inserts actors of various classes into the scene and synthesizes their sizes, orientations, and velocities. We demonstrate on two large-scale datasets SceneGen's ability to faithfully model distributions of real traffic scenes. Moreover, we show that SceneGen coupled with sensor simulation can be used to train perception models that generalize to the real world.

Sourav Biswas, Jerry Liu, Kelvin Wong et al.

We present a novel compression algorithm for reducing the storage of LiDAR sensor data streams. Our model exploits spatio-temporal relationships across multiple LiDAR sweeps to reduce the bitrate of both geometry and intensity values. Towards this goal, we propose a novel conditional entropy model that models the probabilities of the octree symbols by considering both coarse level geometry and previous sweeps' geometric and intensity information. We then use the learned probability to encode the full data stream into a compact one. Our experiments demonstrate that our method significantly reduces the joint geometry and intensity bitrate over prior state-of-the-art LiDAR compression methods, with a reduction of 7-17% and 15-35% on the UrbanCity and SemanticKITTI datasets respectively.

Kelvin Wong, Qiang Zhang, Ming Liang et al.

We present a novel method for testing the safety of self-driving vehicles in simulation. We propose an alternative to sensor simulation, as sensor simulation is expensive and has large domain gaps. Instead, we directly simulate the outputs of the self-driving vehicle's perception and prediction system, enabling realistic motion planning testing. Specifically, we use paired data in the form of ground truth labels and real perception and prediction outputs to train a model that predicts what the online system will produce. Importantly, the inputs to our system consists of high definition maps, bounding boxes, and trajectories, which can be easily sketched by a test engineer in a matter of minutes. This makes our approach a much more scalable solution. Quantitative results on two large-scale datasets demonstrate that we can realistically test motion planning using our simulations.

Sivabalan Manivasagam, Shenlong Wang, Kelvin Wong et al.

We tackle the problem of producing realistic simulations of LiDAR point clouds, the sensor of preference for most self-driving vehicles. We argue that, by leveraging real data, we can simulate the complex world more realistically compared to employing virtual worlds built from CAD/procedural models. Towards this goal, we first build a large catalog of 3D static maps and 3D dynamic objects by driving around several cities with our self-driving fleet. We can then generate scenarios by selecting a scene from our catalog and "virtually" placing the self-driving vehicle (SDV) and a set of dynamic objects from the catalog in plausible locations in the scene. To produce realistic simulations, we develop a novel simulator that captures both the power of physics-based and learning-based simulation. We first utilize ray casting over the 3D scene and then use a deep neural network to produce deviations from the physics-based simulation, producing realistic LiDAR point clouds. We showcase LiDARsim's usefulness for perception algorithms-testing on long-tail events and end-to-end closed-loop evaluation on safety-critical scenarios.

Lila Huang, Shenlong Wang, Kelvin Wong et al.

We present a novel deep compression algorithm to reduce the memory footprint of LiDAR point clouds. Our method exploits the sparsity and structural redundancy between points to reduce the bitrate. Towards this goal, we first encode the LiDAR points into an octree, a data-efficient structure suitable for sparse point clouds. We then design a tree-structured conditional entropy model that models the probabilities of the octree symbols to encode the octree into a compact bitstream. We validate the effectiveness of our method over two large-scale datasets. The results demonstrate that our approach reduces the bitrate by 10-20% at the same reconstruction quality, compared to the previous state-of-the-art. Importantly, we also show that for the same bitrate, our approach outperforms other compression algorithms when performing downstream 3D segmentation and detection tasks using compressed representations. Our algorithm can be used to reduce the onboard and offboard storage of LiDAR points for applications such as self-driving cars, where a single vehicle captures 84 billion points per day

Kelvin Wong, Shenlong Wang, Mengye Ren et al.

In the past few years, we have seen great progress in perception algorithms, particular through the use of deep learning. However, most existing approaches focus on a few categories of interest, which represent only a small fraction of the potential categories that robots need to handle in the real-world. Thus, identifying objects from unknown classes remains a challenging yet crucial task. In this paper, we develop a novel open-set instance segmentation algorithm for point clouds which can segment objects from both known and unknown classes in a holistic way. Our method uses a deep convolutional neural network to project points into a category-agnostic embedding space in which they can be clustered into instances irrespective of their semantics. Experiments on two large-scale self-driving datasets validate the effectiveness of our proposed method.

Yuwen Xiong, Mengye Ren, Renjie Liao et al.

Point clouds are the native output of many real-world 3D sensors. To borrow the success of 2D convolutional network architectures, a majority of popular 3D perception models voxelize the points, which can result in a loss of local geometric details that cannot be recovered. In this paper, we propose a novel learnable convolution layer for processing 3D point cloud data directly. Instead of discretizing points into fixed voxels, we deform our learnable 3D filters to match with the point cloud shape. We propose to combine voxelized backbone networks with our deformable filter layer at 1) the network input stream and 2) the output prediction layers to enhance point level reasoning. We obtain state-of-the-art results on LiDAR semantic segmentation and producing a significant gain in performance on LiDAR object detection.